Lecture 9: Filtration and Martingales (PDF)
Total Page:16
File Type:pdf, Size:1020Kb
MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 9 10/2/2013 Conditional expectations, filtration and martingales Content. 1. Conditional expectations 2. Martingales, sub-martingales and super-martingales 1 Conditional Expectations 1.1 Definition Recall how we define conditional expectations. Given a random variable X and an event A we define [XjA] = E[X1fAg] . E P(A) Also we can consider conditional expectations with respect to random vari ables. For simplicity say Y is a simple random variable on Ω taking values y1; y2; : : : ; yn with some probabilities P(! : Y (!) = yi) = pi: Now we define conditional expectation E[XjY ] as a random variable which takes value E[XjY = yi] with probability pi, where E[XjY = yi] should be understood as expectation of X conditioned on the event f! 2 Ω: Y (!) = yig. It turns out that one can define conditional expectation with respect to a σ field. This notion will include both conditioning on events and conditioning on random variables as special cases. Definition 1. Given Ω, two σ-fields G ⊂ F on Ω, and a probability measure P on (Ω; F). Suppose X is a random variable with respect to F but not necessar ily with respect to G, and suppose X has a finite L1 norm (that is E[jXj] < 1). The conditional expectation E[XjG] is defined to be a random variable Y which satisfies the following properties: (a) Y is measurable with respect to G. 1 (b) For every A 2 G, we have E[X1fAg] = E[Y 1fAg]. For simplicity, from now on we write Z 2 F to indicate that Z is measurable with respect to F. Also let F(Z) denote the smallest σ-field such with respect to which Z is measurable. Theorem 1. The conditional expectation E[XjG] exists and is unique. 0 Uniqueness means that if Y 2 G is any other random variable satisfying 0 conditions (a),(b), then Y = Y a.s. (with respect to measure P). We will prove this theorem using the notion of Radon-Nikodym derivative, the existence of which we state without a proof below. But before we do this, let us develop some intuition behind this definition. 1.2 Simple properties • Consider the trivial case when G = fØ; Ωg. We claim that the constant value c = E[X] is E[XjG]. Indeed, any constant function is measurable with respect to any σ-field So (a) holds. For (b), we have E[X1fΩg] = E[X] = c and E[c1fΩg] = E[c] = c; and E[X1fØg] = 0 and E[c1fØg] = 0. • As the other extreme, suppose G = F. Then we claim that X = E[XjG]. The condition (b) trivially holds. The condition (a) also holds because of the equality between two σ-fields. • Let us go back to our example of conditional expectation with respect to an event A ⊂ Ω. Consider the associated σ-fields G = fØ; A; Ac ; Ωg (we established in the first lecture that this is indeed a σ-field). Consider a random variable Y :Ω ! R defined as E[X1fAg] Y (!) = E[XjA] = , c1 P(A) for ! 2 A and c c E[X1fA g] Y (!) = E[XjA ] = , c2 P(Ac) c for ! 2 A . We claim that Y = E[XjG]. First Y 2 G. Indeed, assume for simplicity c1 < c2. Then f! : Y (!) ≤ xg = Ø when x < c1, = A 2 for c1 ≤ x < c2, = Ω when x ≥ c2. Thus Y 2 G. Then we need to check c equality E[X1fBg] = E[Y 1fBg] for every B = Ø; A; A ; Ω, which is straightforward to do. For example say B = A. Then E[X1fAg] = E[XjA]P(A) = c1P(A): On the other hand we defined Y (!) = c1 for all ! 2 A. Thus E[Y 1fAg] = c1E[1fAg] = c1P(A): And the equality checks. • Suppose now G corresponds to some partition A1;:::;Am of the sample space Ω. Given X 2 F, using a similar analysis, we can check that Y = E[XjG] is a random variable which takes values E[XjAj ] for all ! 2 Aj , for j = 1; 2; : : : ; m. You will recognize that this is one of our earlier examples where we considered conditioning on a simple random variable Y to get E[XjY ]. In fact this generalizes as follows: • Given two random variables X; Y :Ω ! R, suppose both 2 F. Let G = G(Y ) ⊂ F be the field generated by Y . We define E[XjY ] to be E[XjG]. 1.3 Proof of existence We now give a proof sketch of Theorem 1. Proof. Given two probability measures P1; P2 defined on the same (Ω; F), P2 is defined to be absolutely continuous with respect to P1 if for every set A 2 F, P1(A) = 0 implies P2(A) = 0. The following theorem is the main technical part for our proof. It involves using the familiar idea of change of measures. Theorem 2 (Radon-Nikodym Theorem). Suppose P2 is absolutely continuous with respect to P1. Then there exists a non-negative random variable Y :Ω ! R+ such that for every A 2 F P2(A) = EP1 [Y 1fAg]: Function Y is called Radon-Nikodym (RN) derivative and sometimes is denoted dP2=dP1. 3 0 Problem 1. Prove that Y is unique up-to measure zero. That is if Y is also RN 0 derivative, then Y = Y a.s. w.r.t. P1 and hence P2. We now use this theorem to establish the existence of conditional expec tations. Thus we have G ⊂ F, P is a probability measure on F and X is measurable with respect to F. We will only consider the case X ≥ 0 such that E[X] < 1. We also assume that X is not constant, so that E[X] > 0. Consider a new probability measure P2 on G defined as follows: EP[X1fAg] P2(A) = ;A 2 G; EP[X] where we write EP in place of E to emphasize that the expectation operator is with respect to the original measure P. Check that this is indeed a probability measure on (Ω; G). Now P also induced a probability measure on (Ω; G). We claim that P2 is absolutely continuous with respect to P. Indeed if P(A) = 0 then the numerator is zero. By the Radon-Nikodym Theorem then there exists Z which is measurable with respect to G such that for any A 2 G P2(A) = EP[Z1fAg]: We now take Y = ZEP[X]. Then Y satisfies the condition (b) of being a conditional expectation, since for every set B EP[Y 1fBg] = EP[X]EP[Z1fBg] = EP[X1fBg]: The second part, corresponding to the uniqueness property is proved similarly to the uniqueness of the RN derivative (Problem 1). 2 Properties Here are some additional properties of conditional expectations. Linearity. E[aX + Y jG] = aE[XjG] + E[Y jG]. Monotonicity. If X1 ≤ X2 a.s, then E[X1jG] ≤ E[X2jG]. Proof idea is similar to the one you need to use for Problem 1. Independence. Problem 2. Suppose X is independent from G. Namely, for every measurable A ⊂ R;B 2 G P(fX 2 Ag \ B) = P(X 2 A)P(B). Prove that E[XjG] = E[X]. 4 Conditional Jensen’s inequality. Let φ be a convex function and E[jXj]; E[jφ(X)j] < 1. Then φ(E[XjG]) ≤ E[φ(X)jG]. Proof. We use the following representation of a convex function, which we do not prove (see Durrett [1]). Let A = f(a; b) 2 Q : ax + b ≤ φ(x); 8 xg: Then φ(x) = supfax + b :(a; b) 2 Ag. Now we prove the Jensen’s inequality. For any pair of rationals a; b 2 Q satisfying the bound above, we have, by monotonicity that E[φ(X)jG] ≥ aE[XjG] + b, a.s., implying E[φ(X)jG] ≥ supfaE[XjG] + b :(a; b) 2 Ag = φ(E[XjG]) a.s. Tower property. Suppose G1 ⊂ G2 ⊂ F. Then E[E[XjG1]jG2] = E[XjG1] and E[E[XjG2]jG1] = E[XjG1]. That is the smaller field wins. Proof. By definition E[XjG1] is G1 measurable. Therefore it is G2 measurable. Then the first equality follows from the fact E[XjG] = X, when X 2 G, which we established earlier. Now fix any A 2 G1. Denote E[XjG1] by Y1 and E[XjG2] by Y2. Then Y1 2 G1;Y2 2 G2. Then E[Y11fAg] = E[X1fAg]; simply by the definition of Y1 = E[XjG1]. On the other hand, we also have A 2 G2. Therefore E[X1fAg] = E[Y21fAg]: Combining the two equalities we see that E[Y21fAg] = E[Y11fAg] for every A 2 G1. Therefore, E[Y2jG1] = Y1, which is the desired result. An important special case is when G1 is a trivial σ-field fØ; Ωg. We obtain that for every field G E[E[XjG]] = E[X]: 5 3 Filtration and martingales 3.1 Definition A family of σ-fields fFtg is defined to be a filtration if Ft1 ⊂ Ft2 whenever t1 ≤ t2. We will consider only two cases when t 2 Z+ or t 2 R+. A stochastic process fXtg is said to be adapted to filtration fFtg if Xt 2 Ft for every t. Definition 2. A stochastic process fXtg adapted to a filtration fFtg is defined to be a martingale if 1. E[jXtj] < 1 for all t. 2. E[XtjFs] = Xs, for all s < t. When equality is substituted with ≤, the process is called supermartingale. When it is substituted with ≥, the process is called submartingale. Suppose we have a stochastic process fXtg adapted to filtration fFtg and 0 suppose for some s < s < t we have E[XtjFs] = Xs and E[XsjFs0 ] = Xs0 .